light_triad

The biggest challenge is designing the mechanics that make games fun. AI is not quite there yet but with enough data could potentially get there.

I put together a site a few months back to play around with 3d assets from Sketchfab and animated characters from Mixamo as an experiment: https://gobananas.co/

The big problem with education is that all the work is done for grades instead of aiming to deliver value to others. It's important for students to practice but if you raise the bar high enough then AI is just one tool that helps you but doesn't have all the answers. If AI can do all the homework we have to ask ourselves how much value the homework had in the first place.

> Now there's no such thing as a startup, there are only "AI startups", where you build a company that somebody else can write a prompt for in 30 seconds.

The AI hype and AI doom are 2 sides of the same coin. Yes you can make a cool demo quickly with a prompt, but building a real company with moats is still possible, much easier even. The moats have just changed: proprietary data, distribution, access rights, eval sets, integrations, compliance, brand, just not the code.

Some sectors are less productive, but the reality is more mundane: only about 30% of the global workforce are knowledge workers. Adopting AI in real workflows takes time, and that time saved might not be used productively. Also solving Millennium problems is an amazing feat of engineering and mathematics, but doesn't change people's lives that much.

You're right using Microsoft products is like mixing Momento with Waiting for Gadot :)

I recently switched to New Outlook and it's full of weird subtle bugs. It takes them years to fix things and make them barely useable. Given their track record I'm not optimistic for Copilot given both the vision and the execution are questionable at best.

According to this blog post by Anton Labs they are using Sonnet 4.6: https://andonlabs.com/blog/andon-market-launch

> Mr. Petersson and Mr. Backlund said they thought Luna would eventually get smarter and more business-savvy

> having an A.I. boss can be like working for a teddy bear with amnesia.

It makes you question their assumptions when setting this up. Makes sense as PR but it ends up shaping how many readers see AI.

William Thurston's argument (from the quote) was that the point of math isn't to produce more math but to better understand the nature of reality, just like the point of code is not to generate more code but to increase user value.

AI is a tool that reshapes knowledge work but the ultimate goal remains about value.

The terminology can be confusing because in corporate speak being uncompromising can mean being opinionated. Not "we made no tradeoffs" but "we didn't split the difference."

The article means it as tradeoff rather than concession, so the argument is "to avoid design by committee you have to compromise on purpose."

This is the taste question applied to Mathematics in a similar way it's been applied to code. In an era of AI abundance, the question becomes what the goals are and what gets verified and digested (adopted by users). Goodhart’s law: the goal of producing code is not just about maximizing the number of tokens used, but the productivity gains and economic surplus.

This could serve as the template for any field in the age of AI: "We are not trying to meet some abstract production quota. The measure of our success is whether what we do enables people to understand and think more clearly and effectively about math (or products, or science, or hardware...)"

A good example of "founder bias" where big companies are read as not innovating, when in fact their goal is to squeeze as much juice from their user base and strengthen their monopolistic position and pricing power. From the outside it looks like blindness and atrophy but from the inside it's the main bread and butter.

The sighted engineer is also cave adapted, just to a different cave.

The article frames the cost of living issue mainly in terms of demand (...because of OpenAI and Anthropic...), which is an important factor, but has very little to say about supply. San Francisco is a major metropolis that looks like a mining town: there's very few high-rise buildings compared to other major cities. It's due to many factors including strict zoning, growth caps, seismic risks (compared to say Tokyo?) and landlords that don't have much incentive to decrease the value of their skyrocketing assets.

Also some recent setbacks like our leaning Millennium Tower.

It might take a political earthquake to change the status quo given how ossified everything is unfortunately.

Part of the issue is the respective positioning:

- OpenAI wants to be the consumer version of AI, modeled after Google and Meta, with a mostly free universal service powered by ads and e-commerce. They haven't fully shown that model can work. The big problem is the lack of zero marginal costs as each new user requires GPU spend.

- Anthropic positions itself more as enterprise AI, modeled after Microsoft ironically enough, and charges big companies for services. The economics of coding agents work but GPUs get expensive fast and open models are getting good enough for most use cases.

So it's a race between ads and e-commerce offsetting AI spend and open source eating almost everyone's lunch.

Interesting they agreed to sell after their rebrand to Fin a month ago.

There's increasing competition in the customer support AI agent space: Sierra valued at $15.8 billion, Decagon at $4.5 billion. It looks like Salesforce CEO Marc Benioff is trying to compete directly with Sierra, which was started by his ex-Co-CEO Bret Taylor. Also about preventing independent AI support agents from becoming a control point outside the CRM.

I've been using Claude Code regularly since the 4.5 release, and 4.7 was a significant regression: very unreliable, arguing about changes, deciding that fixes weren't needed, etc.

I'm hoping they recreate the magic of 4.5 but it's as much about the quality of harness, the memory and efficiency of the tools than simply the models at this point.

It's a great product. They had the brand, the capital, and the user base to become what Slack, Zoom, or Notion became. Instead, they spent a decade fighting a losing battle over storage pricing with Google and Microsoft. Their lack of a second act is due to a failure of product vision and enterprise execution.

It was both the market and the leadership.

Dropbox failed to find a second act: they struggled to find PMF with their acquisitions and new products: Dropbox Passwords, Dropbox Paper, Carousel etc.

As Steve Jobs warned Drew Houston, Dropbox was "a feature, not a product"

Every generation of builders believed their tools defined their value. Then the tools got easier, faster, automated, and the definition had to change.

But programming didn’t disappear. Writing didn’t disappear. Designing didn’t disappear.

AI flips the equation: when creation becomes cheap, value shifts from how much you can produce to what changes because you showed up. The ability to have a positive impact has actually expanded.

There's still huge gaps in our understanding: quantum gravity, dark matter, what happens before planck time, thermodynamics of life and many others.

Part of the problem is that building bigger colliders, telescopes, and gravitational wave detectors requires huge resources and very powerful computers to store and crunch all the data.

We're cutting research instead of funding it right now and sending our brightest researchers to Europe and China...

Mia was right. Listen to Mia

Here's some context: Tesla, BYD, and Xiaomi Are Playing Different Games https://gilpignol.substack.com/p/tesla-byd-and-xiaomi-are-pl...

Gravity is coming back to Silicon Valley: workers are realizing that the Bell Labs image they were sold was mostly innovation theater and hoarding talent for websites overstuffed with ads designed to manipulate users into buying junk

Distribution, brand, network effects, regulatory positioning, and execution speed all create defensibility; "data helps" doesn't imply "data is everything"

Also as foundation models improve, today's "hard to solve" problems become tomorrow's "easy to solve" problems

It's good for YC to do this and will benefit every startup in the long run. Google has been one of the sources of the AI boom, and provides liquidity by acquiring startups. But as YC argues they've monopolised distribution channels to the point where you need to go through the Google toll booth every time you want to access the market. This tax on founders to reach their audience makes many types of businesses unsustainable and impossible, especially for products where usage != sharing.

The decision can be read in the larger political context. There was some controversy a while back on certain directions the Foundation took like the project on sanitation (aka the toilet challenge) and the backing of charter schools. Regardless of one's opinion of those, he is taking a stand and drawing a line in the sand.

Interesting that some people here are advocating for fewer investments and board seats when angel investing. It's rather the opposite: you need more investments and less time spent on companies.

Angle investors are vital to the startup community as they take a lot of risk and are able to fund ideas that sound too crazy to get regular funding. In this case though given a hit rate of let's say 1% for argument's sake, 54 investments in 15 years is too low. You need hundred of investments to make the math work. Also doing it for fun and to learn is laudable but in the end hitting an Uber or an Airbnb is all that matters. It sounds like the job became full time consulting which is great if you're wealthy and want to give back to the community but can become unsustainable otherwise.

OpenAI has a massive brand advantage because the general public equates their products with AI.

Even if they don't 100% figure out agents they are now big enough that they can acquire those that do.

If the future is mostly about the app layer then they'll be very aggressive in consolidating the same way Facebook did with social media, see for example Windsurf.

AI is changing fast! And to be fair to the model companies, they have been releasing products of (mostly) increasing quality.

It really depends what your use case is. Over the range of all possible use cases this has been the narrative.

I tried Google's model for coding but it kept giving me wrong code. Currently Claude for coding and ChatGPT for more general questions is working for me. The more exotic your use case, the more hit or miss it's going to be.

You sound like you're from the Google team ;)

Rough steps:

1. Monitor keywords

2. Jump in to sway conversation

3. Profit

I'm not saying this is happening. Purely hypothetical.

If you're interested in the functioning of mechanical watches, they're amazing:

https://ciechanow.ski/mechanical-watch/

Previously on HN in 2022: https://news.ycombinator.com/item?id=31261533

This is key: computation is about making things happen. Coding with an LLM adds a level of abstraction but the need for precision and correctness of the "things that happen" doesn't go away. No matter how many cool demos and "coding is dead" pronouncements because AI - and the demos are very cool - the bulk of the work moves to the pre- and post-processing and evals with AI. To the extent that it makes programming more accessible it's a good thing, but can't really replace it.